Hallo community,
I working currently on a vision problem and want to use cuda to fulfill some realtime contraints. To solve the vision problem I want to utilize ROS Kinetic and OpenCV.
My question is now is it possible to replace the opencv version which is delivered with ROS Kinectic (OpenCV3.2.0) with the same opencv version with enabled CUDA?
The most information how to setup and compile the correct opencv version can be found on the ROS wiki page for the opencv3 package I guess (wiki.ros.org/opencv3 section 5. Build the package). Here is described how to checkout the right repos and modify the CMAKE file in the right way. For CUDA I have to add the -DWITH_CUDA flag. Everthing right until now?
Now the complex part: Which software have to be uninstalled before installing the own compiled version and where I have to install, it so ROS can work with it naturally with the libs which are delivered with ROS Kinetic?
My second question is motivated by a statement which I also have found on the wiki page from ROS opencv3 package. There is stated that with opencv 3.2.0 it is now possible to build opencv contrib with installed opencv. This seems interesting for me because I want to modify or better add functionality to a algorithm which is implemented in opencv contrib rgbd.
Is it possible to build only the opencv contrib rgbd package as debug and working on it? At the moment I have no idea how to do it, so will be great when you can help me out.
When you have some question on your side, feel free to ask.
System: Ubuntu 16.04, CUDA8.0, ROS Kinetic, OpenCV 3.2.0
best regards,
Manuel